Comparative Assessment of Risk Score Performance in Ruptured Abdominal Aortic Aneurysm Repair Risk Stratification

Several risk scores are available to stratify the mortality risk of ruptured abdominal aortic aneurysm (rAAA) repair. This study compared the performance of seven risk scores found in the literature using the National Surgical Quality Improvement Program (NSQIP) and Vascular Quality Initiative (VQI) registries.

All patients who underwent open or endovascular rAAA repair in NSQIP (2011 - 2018) and VQI (2003 - 2021) were included. The following risk scores were calculated: Dutch Aneurysm Score (DAS); Glasgow Aneurysm Score (GAS); modified Harborview Risk Score (mHRS); Hardman index (HI); Leiden score; Medicare risk score; and Vascular Surgery Group of New England (VSGNE) risk score. Discrimination was assessed for all patients (combined open and endovascular) using receiver operating characteristic (ROC) curves, with area under the curve (AUC) values compared within datasets using the DeLong test. Calibration was evaluated using graphical calibration curves and was quantified via the integrated calibration index (ICI).

A total of 2 134 NSQIP patients and 6 458 VQI patients were included. In NSQIP, the GAS had the highest discrimination ability (AUC 0.687; p < .010 vs. all); while in VQI, the DAS performed best (AUC 0.688; p < .050 vs. all). The Medicare risk score exhibited the best calibration with an ICI of 0.007 in NSQIP and 0.003 in VQI. The DAS tended to underestimate risk, while all other scores overestimated death.

Risk scores serve as valuable tools to augment, but not replace, clinical decision making in patients with rAAA. The GAS and DAS demonstrated superior discrimination, rendering them best for distinguishing between low and high risk patients. Meanwhile, the Medicare risk score offers the most accurate risk prediction. The mHRS and HI offered a balance of adequate discrimination and calibration while remaining simple and easily calculable, making them practical for urgent clinical settings.